Merge #5246

5246: Fix dump import r=Kerollmops a=dureuill

- Fix: handle the change of format of the update files
  - Correctly handle update files as JSON stream rather than obkv when exporting a dump with enqueued tasks
  - Correctly recreate update files as JSON stream rather than obkv when importing a dump
  - As the dump format itself didn't change, all dumps are still compatible
- Temporary workaround for https://github.com/meilisearch/meilisearch/issues/5247: set the batch uid of tasks to `null` at dump export time.
- Changes to meilitool
  - Export dump with update files in new format if DB >= v1.12
  - offline upgrade now supports upgrading from [1.9.0-1.12.5] to [1.10.0-1.12.5].
  - offline upgrade supports no-op upgrades and has better error messages 

Co-authored-by: Louis Dureuil <louis@meilisearch.com>
Co-authored-by: ManyTheFish <many@meilisearch.com>

Cargo.lock

@@ -496,7 +496,7 @@ source = "git+https://github.com/meilisearch/bbqueue#cbb87cc707b5af415ef203bdaf2
 
 [[package]]
 name = "benchmarks"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "bumpalo",
@@ -689,7 +689,7 @@ dependencies = [
 
 [[package]]
 name = "build-info"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "time",
@@ -1664,7 +1664,7 @@ dependencies = [
 
 [[package]]
 name = "dump"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "big_s",
@@ -1876,7 +1876,7 @@ checksum = "486f806e73c5707928240ddc295403b1b93c96a02038563881c4a2fd84b81ac4"
 
 [[package]]
 name = "file-store"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "tempfile",
  "thiserror",
@@ -1898,7 +1898,7 @@ dependencies = [
 
 [[package]]
 name = "filter-parser"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "insta",
  "nom",
@@ -1918,7 +1918,7 @@ dependencies = [
 
 [[package]]
 name = "flatten-serde-json"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "criterion",
  "serde_json",
@@ -2057,7 +2057,7 @@ dependencies = [
 
 [[package]]
 name = "fuzzers"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "arbitrary",
  "bumpalo",
@@ -2624,7 +2624,7 @@ checksum = "206ca75c9c03ba3d4ace2460e57b189f39f43de612c2f85836e65c929701bb2d"
 
 [[package]]
 name = "index-scheduler"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "arroy 0.5.0 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -2822,7 +2822,7 @@ dependencies = [
 
 [[package]]
 name = "json-depth-checker"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "criterion",
  "serde_json",
@@ -3441,7 +3441,7 @@ checksum = "490cc448043f947bae3cbee9c203358d62dbee0db12107a74be5c30ccfd09771"
 
 [[package]]
 name = "meili-snap"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "insta",
  "md5",
@@ -3450,7 +3450,7 @@ dependencies = [
 
 [[package]]
 name = "meilisearch"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "actix-cors",
  "actix-http",
@@ -3540,7 +3540,7 @@ dependencies = [
 
 [[package]]
 name = "meilisearch-auth"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "base64 0.22.1",
  "enum-iterator",
@@ -3559,7 +3559,7 @@ dependencies = [
 
 [[package]]
 name = "meilisearch-types"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "actix-web",
  "anyhow",
@@ -3592,7 +3592,7 @@ dependencies = [
 
 [[package]]
 name = "meilitool"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "arroy 0.5.0 (git+https://github.com/meilisearch/arroy/?tag=DO-NOT-DELETE-upgrade-v04-to-v05)",
@@ -3627,7 +3627,7 @@ dependencies = [
 
 [[package]]
 name = "milli"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "allocator-api2",
  "arroy 0.5.0 (registry+https://github.com/rust-lang/crates.io-index)",
@@ -4083,7 +4083,7 @@ checksum = "e3148f5046208a5d56bcfc03053e3ca6334e51da8dfb19b6cdc8b306fae3283e"
 
 [[package]]
 name = "permissive-json-pointer"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "big_s",
  "serde_json",
@@ -6486,7 +6486,7 @@ dependencies = [
 
 [[package]]
 name = "xtask"
-version = "1.12.0"
+version = "1.12.5"
 dependencies = [
  "anyhow",
  "build-info",

Cargo.toml

@@ -22,7 +22,7 @@ members = [
 ]
 
 [workspace.package]
-version = "1.12.0"
+version = "1.12.5"
 authors = [
     "Quentin de Quelen <quentin@dequelen.me>",
     "Clément Renault <clement@meilisearch.com>",

crates/index-scheduler/src/batch.rs

@@ -29,7 +29,7 @@ use bumpalo::Bump;
 use dump::IndexMetadata;
 use meilisearch_types::batches::BatchId;
 use meilisearch_types::heed::{RoTxn, RwTxn};
-use meilisearch_types::milli::documents::{obkv_to_object, DocumentsBatchReader, PrimaryKey};
+use meilisearch_types::milli::documents::PrimaryKey;
 use meilisearch_types::milli::heed::CompactionOption;
 use meilisearch_types::milli::progress::Progress;
 use meilisearch_types::milli::update::new::indexer::{self, UpdateByFunction};
@@ -819,6 +819,13 @@ impl IndexScheduler {
                         t.started_at = Some(started_at);
                         t.finished_at = Some(finished_at);
                     }
+
+                    // Patch the task to remove the batch uid, because as of v1.12.5 batches are not persisted.
+                    // This prevent from referencing *future* batches not actually associated with the task.
+                    //
+                    // See <https://github.com/meilisearch/meilisearch/issues/5247> for details.
+                    t.batch_uid = None;
+
                     let mut dump_content_file = dump_tasks.push_task(&t.into())?;
 
                     // 2.1. Dump the `content_file` associated with the task if there is one and the task is not finished yet.
@@ -829,21 +836,18 @@ impl IndexScheduler {
                         if status == Status::Enqueued {
                             let content_file = self.file_store.get_update(content_file)?;
 
-                            let reader = DocumentsBatchReader::from_reader(content_file)
+                            for document in
-                                .map_err(|e| Error::from_milli(e.into(), None))?;
+                                serde_json::de::Deserializer::from_reader(content_file).into_iter()
-
-                            let (mut cursor, documents_batch_index) =
-                                reader.into_cursor_and_fields_index();
-
-                            while let Some(doc) = cursor
-                                .next_document()
-                                .map_err(|e| Error::from_milli(e.into(), None))?
                             {
-                                dump_content_file.push_document(
+                                let document = document.map_err(|e| {
-                                    &obkv_to_object(doc, &documents_batch_index)
+                                    Error::from_milli(
-                                        .map_err(|e| Error::from_milli(e, None))?,
+                                        milli::InternalError::SerdeJson(e).into(),
-                                )?;
+                                        None,
+                                    )
+                                })?;
+                                dump_content_file.push_document(&document)?;
                             }
+
                             dump_content_file.flush()?;
                         }
                     }
@@ -1312,9 +1316,7 @@ impl IndexScheduler {
                     if let DocumentOperation::Add(content_uuid) = operation {
                         let content_file = self.file_store.get_update(*content_uuid)?;
                         let mmap = unsafe { memmap2::Mmap::map(&content_file)? };
-                        if !mmap.is_empty() {
+                        content_files.push(mmap);
-                            content_files.push(mmap);
-                        }
                     }
                 }
 

crates/index-scheduler/src/lib.rs

@@ -55,7 +55,6 @@ use meilisearch_types::features::{InstanceTogglableFeatures, RuntimeTogglableFea
 use meilisearch_types::heed::byteorder::BE;
 use meilisearch_types::heed::types::{SerdeBincode, SerdeJson, Str, I128};
 use meilisearch_types::heed::{self, Database, Env, PutFlags, RoTxn, RwTxn};
-use meilisearch_types::milli::documents::DocumentsBatchBuilder;
 use meilisearch_types::milli::index::IndexEmbeddingConfig;
 use meilisearch_types::milli::update::IndexerConfig;
 use meilisearch_types::milli::vector::{Embedder, EmbedderOptions, EmbeddingConfigs};
@@ -2017,14 +2016,19 @@ impl<'a> Dump<'a> {
         task: TaskDump,
         content_file: Option<Box<UpdateFile>>,
     ) -> Result<Task> {
+        let task_has_no_docs = matches!(task.kind, KindDump::DocumentImport { documents_count, .. } if documents_count == 0);
+
         let content_uuid = match content_file {
             Some(content_file) if task.status == Status::Enqueued => {
-                let (uuid, mut file) = self.index_scheduler.create_update_file(false)?;
+                let (uuid, file) = self.index_scheduler.create_update_file(false)?;
-                let mut builder = DocumentsBatchBuilder::new(&mut file);
+                let mut writer = io::BufWriter::new(file);
                 for doc in content_file {
-                    builder.append_json_object(&doc?)?;
+                    let doc = doc?;
+                    serde_json::to_writer(&mut writer, &doc).map_err(|e| {
+                        Error::from_milli(milli::InternalError::SerdeJson(e).into(), None)
+                    })?;
                 }
-                builder.into_inner()?;
+                let file = writer.into_inner().map_err(|e| e.into_error())?;
                 file.persist()?;
 
                 Some(uuid)
@@ -2032,6 +2036,12 @@ impl<'a> Dump<'a> {
             // If the task isn't `Enqueued` then just generate a recognisable `Uuid`
             // in case we try to open it later.
             _ if task.status != Status::Enqueued => Some(Uuid::nil()),
+            None if task.status == Status::Enqueued && task_has_no_docs => {
+                let (uuid, file) = self.index_scheduler.create_update_file(false)?;
+                file.persist()?;
+
+                Some(uuid)
+            }
             _ => None,
         };
 

crates/index-scheduler/src/utils.rs

@@ -291,7 +291,10 @@ impl IndexScheduler {
 
         debug_assert!(old_task != *task);
         debug_assert_eq!(old_task.uid, task.uid);
-        debug_assert!(old_task.batch_uid.is_none() && task.batch_uid.is_some());
+        debug_assert!(
+            old_task.batch_uid.is_none() && task.batch_uid.is_some(),
+            "\n==> old: {old_task:?}\n==> new: {task:?}"
+        );
 
         if old_task.status != task.status {
             self.update_status(wtxn, old_task.status, |bitmap| {

crates/meilisearch/tests/documents/add_documents.rs

@@ -1220,9 +1220,89 @@ async fn replace_document() {
 #[actix_rt::test]
 async fn add_no_documents() {
     let server = Server::new().await;
-    let index = server.index("test");
+    let index = server.index("kefir");
-    let (_response, code) = index.add_documents(json!([]), None).await;
+    let (task, code) = index.add_documents(json!([]), None).await;
     snapshot!(code, @"202 Accepted");
+    let task = server.wait_task(task.uid()).await;
+    let task = task.succeeded();
+    snapshot!(task, @r#"
+    {
+      "uid": "[uid]",
+      "batchUid": "[batch_uid]",
+      "indexUid": "kefir",
+      "status": "succeeded",
+      "type": "documentAdditionOrUpdate",
+      "canceledBy": null,
+      "details": {
+        "receivedDocuments": 0,
+        "indexedDocuments": 0
+      },
+      "error": null,
+      "duration": "[duration]",
+      "enqueuedAt": "[date]",
+      "startedAt": "[date]",
+      "finishedAt": "[date]"
+    }
+    "#);
+
+    let (task, _code) = index.add_documents(json!([]), Some("kefkef")).await;
+    let task = server.wait_task(task.uid()).await;
+    let task = task.succeeded();
+    snapshot!(task, @r#"
+    {
+      "uid": "[uid]",
+      "batchUid": "[batch_uid]",
+      "indexUid": "kefir",
+      "status": "succeeded",
+      "type": "documentAdditionOrUpdate",
+      "canceledBy": null,
+      "details": {
+        "receivedDocuments": 0,
+        "indexedDocuments": 0
+      },
+      "error": null,
+      "duration": "[duration]",
+      "enqueuedAt": "[date]",
+      "startedAt": "[date]",
+      "finishedAt": "[date]"
+    }
+    "#);
+
+    let (task, _code) = index.add_documents(json!([{ "kefkef": 1 }]), None).await;
+    let task = server.wait_task(task.uid()).await;
+    let task = task.succeeded();
+    snapshot!(task, @r#"
+    {
+      "uid": "[uid]",
+      "batchUid": "[batch_uid]",
+      "indexUid": "kefir",
+      "status": "succeeded",
+      "type": "documentAdditionOrUpdate",
+      "canceledBy": null,
+      "details": {
+        "receivedDocuments": 1,
+        "indexedDocuments": 1
+      },
+      "error": null,
+      "duration": "[duration]",
+      "enqueuedAt": "[date]",
+      "startedAt": "[date]",
+      "finishedAt": "[date]"
+    }
+    "#);
+    let (documents, _status) = index.get_all_documents(GetAllDocumentsOptions::default()).await;
+    snapshot!(documents, @r#"
+    {
+      "results": [
+        {
+          "kefkef": 1
+        }
+      ],
+      "offset": 0,
+      "limit": 20,
+      "total": 1
+    }
+    "#);
 }
 
 #[actix_rt::test]

crates/meilitool/Cargo.toml

@@ -14,11 +14,11 @@ arroy_v04_to_v05 = { package = "arroy", git = "https://github.com/meilisearch/ar
 clap = { version = "4.5.9", features = ["derive"] }
 dump = { path = "../dump" }
 file-store = { path = "../file-store" }
-indexmap = {version = "2.7.0", features = ["serde"]}
+indexmap = { version = "2.7.0", features = ["serde"] }
 meilisearch-auth = { path = "../meilisearch-auth" }
 meilisearch-types = { path = "../meilisearch-types" }
 serde = { version = "1.0.209", features = ["derive"] }
-serde_json = {version = "1.0.133", features = ["preserve_order"]}
+serde_json = { version = "1.0.133", features = ["preserve_order"] }
 tempfile = "3.14.0"
 time = { version = "0.3.36", features = ["formatting", "parsing", "alloc"] }
 uuid = { version = "1.10.0", features = ["v4"], default-features = false }

crates/meilitool/src/main.rs

@@ -88,7 +88,7 @@ fn main() -> anyhow::Result<()> {
     match command {
         Command::ClearTaskQueue => clear_task_queue(db_path),
         Command::ExportADump { dump_dir, skip_enqueued_tasks } => {
-            export_a_dump(db_path, dump_dir, skip_enqueued_tasks)
+            export_a_dump(db_path, dump_dir, skip_enqueued_tasks, detected_version)
         }
         Command::OfflineUpgrade { target_version } => {
             let target_version = parse_version(&target_version).context("While parsing `--target-version`. Make sure `--target-version` is in the format MAJOR.MINOR.PATCH")?;
@@ -187,6 +187,7 @@ fn export_a_dump(
     db_path: PathBuf,
     dump_dir: PathBuf,
     skip_enqueued_tasks: bool,
+    detected_version: (String, String, String),
 ) -> Result<(), anyhow::Error> {
     let started_at = OffsetDateTime::now_utc();
 
@@ -238,9 +239,6 @@ fn export_a_dump(
     if skip_enqueued_tasks {
         eprintln!("Skip dumping the enqueued tasks...");
     } else {
-        eprintln!("Dumping the enqueued tasks...");
-
-        // 3. dump the tasks
         let mut dump_tasks = dump.create_tasks_queue()?;
         let mut count = 0;
         for ret in all_tasks.iter(&rtxn)? {
@@ -254,18 +252,39 @@ fn export_a_dump(
                 if status == Status::Enqueued {
                     let content_file = file_store.get_update(content_file_uuid)?;
 
-                    let reader =
+                    if (
-                        DocumentsBatchReader::from_reader(content_file).with_context(|| {
+                        detected_version.0.as_str(),
-                            format!("While reading content file {:?}", content_file_uuid)
+                        detected_version.1.as_str(),
-                        })?;
+                        detected_version.2.as_str(),
-
+                    ) < ("1", "12", "0")
-                    let (mut cursor, documents_batch_index) = reader.into_cursor_and_fields_index();
+                    {
-                    while let Some(doc) = cursor.next_document().with_context(|| {
+                        eprintln!("Dumping the enqueued tasks reading them in obkv format...");
-                        format!("While iterating on content file {:?}", content_file_uuid)
+                        let reader =
-                    })? {
+                            DocumentsBatchReader::from_reader(content_file).with_context(|| {
-                        dump_content_file
+                                format!("While reading content file {:?}", content_file_uuid)
-                            .push_document(&obkv_to_object(doc, &documents_batch_index)?)?;
+                            })?;
+                        let (mut cursor, documents_batch_index) =
+                            reader.into_cursor_and_fields_index();
+                        while let Some(doc) = cursor.next_document().with_context(|| {
+                            format!("While iterating on content file {:?}", content_file_uuid)
+                        })? {
+                            dump_content_file
+                                .push_document(&obkv_to_object(doc, &documents_batch_index)?)?;
+                        }
+                    } else {
+                        eprintln!(
+                            "Dumping the enqueued tasks reading them in JSON stream format..."
+                        );
+                        for document in
+                            serde_json::de::Deserializer::from_reader(content_file).into_iter()
+                        {
+                            let document = document.with_context(|| {
+                                format!("While reading content file {:?}", content_file_uuid)
+                            })?;
+                            dump_content_file.push_document(&document)?;
+                        }
                     }
+
                     dump_content_file.flush()?;
                     count += 1;
                 }

crates/meilitool/src/upgrade/mod.rs

@@ -8,7 +8,7 @@ use std::path::{Path, PathBuf};
 use anyhow::{bail, Context};
 use meilisearch_types::versioning::create_version_file;
 use v1_10::v1_9_to_v1_10;
-use v1_12::v1_11_to_v1_12;
+use v1_12::{v1_11_to_v1_12, v1_12_to_v1_12_3};
 
 use crate::upgrade::v1_11::v1_10_to_v1_11;
 
@@ -20,12 +20,48 @@ pub struct OfflineUpgrade {
 
 impl OfflineUpgrade {
     pub fn upgrade(self) -> anyhow::Result<()> {
+        // Adding a version?
+        //
+        // 1. Update the LAST_SUPPORTED_UPGRADE_FROM_VERSION and LAST_SUPPORTED_UPGRADE_TO_VERSION.
+        // 2. Add new version to the upgrade list if necessary
+        // 3. Use `no_upgrade` as index for versions that are compatible.
+
+        if self.current_version == self.target_version {
+            println!("Database is already at the target version. Exiting.");
+            return Ok(());
+        }
+
+        if self.current_version > self.target_version {
+            bail!(
+                "Cannot downgrade from {}.{}.{} to {}.{}.{}. Downgrade not supported",
+                self.current_version.0,
+                self.current_version.1,
+                self.current_version.2,
+                self.target_version.0,
+                self.target_version.1,
+                self.target_version.2
+            );
+        }
+
+        const FIRST_SUPPORTED_UPGRADE_FROM_VERSION: &str = "1.9.0";
+        const LAST_SUPPORTED_UPGRADE_FROM_VERSION: &str = "1.12.5";
+        const FIRST_SUPPORTED_UPGRADE_TO_VERSION: &str = "1.10.0";
+        const LAST_SUPPORTED_UPGRADE_TO_VERSION: &str = "1.12.5";
+
         let upgrade_list = [
-            (v1_9_to_v1_10 as fn(&Path) -> Result<(), anyhow::Error>, "1", "10", "0"),
+            (
+                v1_9_to_v1_10 as fn(&Path, &str, &str, &str) -> Result<(), anyhow::Error>,
+                "1",
+                "10",
+                "0",
+            ),
             (v1_10_to_v1_11, "1", "11", "0"),
             (v1_11_to_v1_12, "1", "12", "0"),
+            (v1_12_to_v1_12_3, "1", "12", "3"),
         ];
 
+        let no_upgrade: usize = upgrade_list.len();
+
         let (current_major, current_minor, current_patch) = &self.current_version;
 
         let start_at = match (
@@ -36,8 +72,12 @@ impl OfflineUpgrade {
             ("1", "9", _) => 0,
             ("1", "10", _) => 1,
             ("1", "11", _) => 2,
+            ("1", "12", "0" | "1" | "2") => 3,
+            ("1", "12", "3" | "4" | "5") => no_upgrade,
             _ => {
-                bail!("Unsupported current version {current_major}.{current_minor}.{current_patch}. Can only upgrade from v1.9 and v1.10")
+                bail!("Unsupported current version {current_major}.{current_minor}.{current_patch}. Can only upgrade from versions in range [{}-{}]",
+                      FIRST_SUPPORTED_UPGRADE_FROM_VERSION,
+                      LAST_SUPPORTED_UPGRADE_FROM_VERSION);
             }
         };
 
@@ -46,21 +86,32 @@ impl OfflineUpgrade {
         let ends_at = match (target_major.as_str(), target_minor.as_str(), target_patch.as_str()) {
             ("1", "10", _) => 0,
             ("1", "11", _) => 1,
-            ("1", "12", _) => 2,
+            ("1", "12", "0" | "1" | "2") => 2,
+            ("1", "12", "3" | "4" | "5") => 3,
             (major, _, _) if major.starts_with('v') => {
                 bail!("Target version must not starts with a `v`. Instead of writing `v1.9.0` write `1.9.0` for example.")
             }
             _ => {
-                bail!("Unsupported target version {target_major}.{target_minor}.{target_patch}. Can only upgrade to v1.10 and v1.11")
+                bail!("Unsupported target version {target_major}.{target_minor}.{target_patch}. Can only upgrade to versions in range [{}-{}]",
+                      FIRST_SUPPORTED_UPGRADE_TO_VERSION,
+                      LAST_SUPPORTED_UPGRADE_TO_VERSION);
             }
         };
 
         println!("Starting the upgrade from {current_major}.{current_minor}.{current_patch} to {target_major}.{target_minor}.{target_patch}");
 
+        if start_at == no_upgrade {
+            println!("No upgrade operation to perform, writing VERSION file");
+            create_version_file(&self.db_path, target_major, target_minor, target_patch)
+                .context("while writing VERSION file after the upgrade")?;
+            println!("Success");
+            return Ok(());
+        }
+
         #[allow(clippy::needless_range_loop)]
         for index in start_at..=ends_at {
             let (func, major, minor, patch) = upgrade_list[index];
-            (func)(&self.db_path)?;
+            (func)(&self.db_path, current_major, current_minor, current_patch)?;
             println!("Done");
             // We're writing the version file just in case an issue arise _while_ upgrading.
             // We don't want the DB to fail in an unknown state.

crates/meilitool/src/upgrade/v1_10.rs

@@ -151,7 +151,12 @@ fn date_round_trip(
     Ok(())
 }
 
-pub fn v1_9_to_v1_10(db_path: &Path) -> anyhow::Result<()> {
+pub fn v1_9_to_v1_10(
+    db_path: &Path,
+    _origin_major: &str,
+    _origin_minor: &str,
+    _origin_patch: &str,
+) -> anyhow::Result<()> {
     println!("Upgrading from v1.9.0 to v1.10.0");
     // 2 changes here
 

crates/meilitool/src/upgrade/v1_11.rs

@@ -14,7 +14,12 @@ use meilisearch_types::milli::index::db_name;
 use crate::uuid_codec::UuidCodec;
 use crate::{try_opening_database, try_opening_poly_database};
 
-pub fn v1_10_to_v1_11(db_path: &Path) -> anyhow::Result<()> {
+pub fn v1_10_to_v1_11(
+    db_path: &Path,
+    _origin_major: &str,
+    _origin_minor: &str,
+    _origin_patch: &str,
+) -> anyhow::Result<()> {
     println!("Upgrading from v1.10.0 to v1.11.0");
 
     let index_scheduler_path = db_path.join("tasks");

crates/meilitool/src/upgrade/v1_12.rs

@@ -1,17 +1,34 @@
 //! The breaking changes that happened between the v1.11 and the v1.12 are:
 //! - The new indexer changed the update files format from OBKV to ndjson. https://github.com/meilisearch/meilisearch/pull/4900
 
+use std::borrow::Cow;
 use std::io::BufWriter;
 use std::path::Path;
+use std::sync::atomic::AtomicBool;
 
 use anyhow::Context;
 use file_store::FileStore;
 use indexmap::IndexMap;
 use meilisearch_types::milli::documents::DocumentsBatchReader;
+use meilisearch_types::milli::heed::types::{SerdeJson, Str};
+use meilisearch_types::milli::heed::{Database, EnvOpenOptions, RoTxn, RwTxn};
+use meilisearch_types::milli::progress::Step;
+use meilisearch_types::milli::{FieldDistribution, Index};
+use serde::Serialize;
 use serde_json::value::RawValue;
 use tempfile::NamedTempFile;
+use time::OffsetDateTime;
+use uuid::Uuid;
 
-pub fn v1_11_to_v1_12(db_path: &Path) -> anyhow::Result<()> {
+use crate::try_opening_database;
+use crate::uuid_codec::UuidCodec;
+
+pub fn v1_11_to_v1_12(
+    db_path: &Path,
+    _origin_major: &str,
+    _origin_minor: &str,
+    _origin_patch: &str,
+) -> anyhow::Result<()> {
     println!("Upgrading from v1.11.0 to v1.12.0");
 
     convert_update_files(db_path)?;
@@ -19,6 +36,23 @@ pub fn v1_11_to_v1_12(db_path: &Path) -> anyhow::Result<()> {
     Ok(())
 }
 
+pub fn v1_12_to_v1_12_3(
+    db_path: &Path,
+    origin_major: &str,
+    origin_minor: &str,
+    origin_patch: &str,
+) -> anyhow::Result<()> {
+    println!("Upgrading from v1.12.{{0, 1, 2}} to v1.12.3");
+
+    if origin_minor == "12" {
+        rebuild_field_distribution(db_path)?;
+    } else {
+        println!("Not rebuilding field distribution as it wasn't corrupted coming from v{origin_major}.{origin_minor}.{origin_patch}");
+    }
+
+    Ok(())
+}
+
 /// Convert the update files from OBKV to ndjson format.
 ///
 /// 1) List all the update files using the file store.
@@ -77,3 +111,188 @@ fn convert_update_files(db_path: &Path) -> anyhow::Result<()> {
 
     Ok(())
 }
+
+/// Rebuild field distribution as it was wrongly computed in v1.12.x if x < 3
+fn rebuild_field_distribution(db_path: &Path) -> anyhow::Result<()> {
+    let index_scheduler_path = db_path.join("tasks");
+    let env = unsafe { EnvOpenOptions::new().max_dbs(100).open(&index_scheduler_path) }
+        .with_context(|| format!("While trying to open {:?}", index_scheduler_path.display()))?;
+
+    let mut sched_wtxn = env.write_txn()?;
+
+    let index_mapping: Database<Str, UuidCodec> =
+        try_opening_database(&env, &sched_wtxn, "index-mapping")?;
+    let stats_db: Database<UuidCodec, SerdeJson<IndexStats>> =
+        try_opening_database(&env, &sched_wtxn, "index-stats").with_context(|| {
+            format!("While trying to open {:?}", index_scheduler_path.display())
+        })?;
+
+    let index_count =
+        index_mapping.len(&sched_wtxn).context("while reading the number of indexes")?;
+
+    // FIXME: not ideal, we have to pre-populate all indexes to prevent double borrow of sched_wtxn
+    // 1. immutably for the iteration
+    // 2. mutably for updating index stats
+    let indexes: Vec<_> = index_mapping
+        .iter(&sched_wtxn)?
+        .map(|res| res.map(|(uid, uuid)| (uid.to_owned(), uuid)))
+        .collect();
+
+    let progress = meilisearch_types::milli::progress::Progress::default();
+    let finished = AtomicBool::new(false);
+
+    std::thread::scope(|scope| {
+        let display_progress = std::thread::Builder::new()
+            .name("display_progress".into())
+            .spawn_scoped(scope, || {
+                while !finished.load(std::sync::atomic::Ordering::Relaxed) {
+                    std::thread::sleep(std::time::Duration::from_secs(5));
+                    let view = progress.as_progress_view();
+                    let Ok(view) = serde_json::to_string(&view) else {
+                        continue;
+                    };
+                    println!("{view}");
+                }
+            })
+            .unwrap();
+
+        for (index_index, result) in indexes.into_iter().enumerate() {
+            let (uid, uuid) = result?;
+            progress.update_progress(VariableNameStep::new(
+                &uid,
+                index_index as u32,
+                index_count as u32,
+            ));
+            let index_path = db_path.join("indexes").join(uuid.to_string());
+
+            println!(
+                "[{}/{index_count}]Updating index `{uid}` at `{}`",
+                index_index + 1,
+                index_path.display()
+            );
+
+            println!("\t- Rebuilding field distribution");
+
+            let index = meilisearch_types::milli::Index::new(EnvOpenOptions::new(), &index_path)
+                .with_context(|| {
+                    format!("while opening index {uid} at '{}'", index_path.display())
+                })?;
+
+            let mut index_txn = index.write_txn()?;
+
+            meilisearch_types::milli::update::new::reindex::field_distribution(
+                &index,
+                &mut index_txn,
+                &progress,
+            )
+            .context("while rebuilding field distribution")?;
+
+            let stats = IndexStats::new(&index, &index_txn)
+                .with_context(|| format!("computing stats for index `{uid}`"))?;
+            store_stats_of(stats_db, uuid, &mut sched_wtxn, &uid, &stats)?;
+
+            index_txn.commit().context("while committing the write txn for the updated index")?;
+        }
+
+        sched_wtxn.commit().context("while committing the write txn for the index-scheduler")?;
+
+        finished.store(true, std::sync::atomic::Ordering::Relaxed);
+
+        if let Err(panic) = display_progress.join() {
+            let msg = match panic.downcast_ref::<&'static str>() {
+                Some(s) => *s,
+                None => match panic.downcast_ref::<String>() {
+                    Some(s) => &s[..],
+                    None => "Box<dyn Any>",
+                },
+            };
+            eprintln!("WARN: the display thread panicked with {msg}");
+        }
+
+        println!("Upgrading database succeeded");
+        Ok(())
+    })
+}
+
+pub struct VariableNameStep {
+    name: String,
+    current: u32,
+    total: u32,
+}
+
+impl VariableNameStep {
+    pub fn new(name: impl Into<String>, current: u32, total: u32) -> Self {
+        Self { name: name.into(), current, total }
+    }
+}
+
+impl Step for VariableNameStep {
+    fn name(&self) -> Cow<'static, str> {
+        self.name.clone().into()
+    }
+
+    fn current(&self) -> u32 {
+        self.current
+    }
+
+    fn total(&self) -> u32 {
+        self.total
+    }
+}
+
+pub fn store_stats_of(
+    stats_db: Database<UuidCodec, SerdeJson<IndexStats>>,
+    index_uuid: Uuid,
+    sched_wtxn: &mut RwTxn,
+    index_uid: &str,
+    stats: &IndexStats,
+) -> anyhow::Result<()> {
+    stats_db
+        .put(sched_wtxn, &index_uuid, stats)
+        .with_context(|| format!("storing stats for index `{index_uid}`"))?;
+    Ok(())
+}
+
+/// The statistics that can be computed from an `Index` object.
+#[derive(Serialize, Debug)]
+pub struct IndexStats {
+    /// Number of documents in the index.
+    pub number_of_documents: u64,
+    /// Size taken up by the index' DB, in bytes.
+    ///
+    /// This includes the size taken by both the used and free pages of the DB, and as the free pages
+    /// are not returned to the disk after a deletion, this number is typically larger than
+    /// `used_database_size` that only includes the size of the used pages.
+    pub database_size: u64,
+    /// Size taken by the used pages of the index' DB, in bytes.
+    ///
+    /// As the DB backend does not return to the disk the pages that are not currently used by the DB,
+    /// this value is typically smaller than `database_size`.
+    pub used_database_size: u64,
+    /// Association of every field name with the number of times it occurs in the documents.
+    pub field_distribution: FieldDistribution,
+    /// Creation date of the index.
+    #[serde(with = "time::serde::rfc3339")]
+    pub created_at: OffsetDateTime,
+    /// Date of the last update of the index.
+    #[serde(with = "time::serde::rfc3339")]
+    pub updated_at: OffsetDateTime,
+}
+
+impl IndexStats {
+    /// Compute the stats of an index
+    ///
+    /// # Parameters
+    ///
+    /// - rtxn: a RO transaction for the index, obtained from `Index::read_txn()`.
+    pub fn new(index: &Index, rtxn: &RoTxn) -> meilisearch_types::milli::Result<Self> {
+        Ok(IndexStats {
+            number_of_documents: index.number_of_documents(rtxn)?,
+            database_size: index.on_disk_size()?,
+            used_database_size: index.used_size()?,
+            field_distribution: index.field_distribution(rtxn)?,
+            created_at: index.created_at(rtxn)?,
+            updated_at: index.updated_at(rtxn)?,
+        })
+    }
+}

crates/milli/src/update/facet/mod.rs

@@ -79,22 +79,29 @@ pub const FACET_MIN_LEVEL_SIZE: u8 = 5;
 use std::collections::BTreeSet;
 use std::fs::File;
 use std::io::BufReader;
+use std::ops::Bound;
 
 use grenad::Merger;
 use heed::types::{Bytes, DecodeIgnore};
+use heed::BytesDecode as _;
+use roaring::RoaringBitmap;
 use time::OffsetDateTime;
 use tracing::debug;
 
 use self::incremental::FacetsUpdateIncremental;
 use super::{FacetsUpdateBulk, MergeDeladdBtreesetString, MergeDeladdCboRoaringBitmaps};
 use crate::facet::FacetType;
-use crate::heed_codec::facet::{FacetGroupKey, FacetGroupKeyCodec, FacetGroupValueCodec};
+use crate::heed_codec::facet::{
+    FacetGroupKey, FacetGroupKeyCodec, FacetGroupValueCodec, OrderedF64Codec,
+};
 use crate::heed_codec::BytesRefCodec;
+use crate::search::facet::get_highest_level;
 use crate::update::del_add::{DelAdd, KvReaderDelAdd};
 use crate::{try_split_array_at, FieldId, Index, Result};
 
 pub mod bulk;
 pub mod incremental;
+pub mod new_incremental;
 
 /// A builder used to add new elements to the `facet_id_string_docids` or `facet_id_f64_docids` databases.
 ///
@@ -646,3 +653,194 @@ mod comparison_bench {
         }
     }
 }
+
+/// Run sanity checks on the specified fid tree
+///
+/// 1. No "orphan" child value, any child value has a parent
+/// 2. Any docid in the child appears in the parent
+/// 3. No docid in the parent is missing from all its children
+/// 4. no group is bigger than max_group_size
+/// 5. Less than 50% of groups are bigger than group_size
+/// 6. group size matches the number of children
+/// 7. max_level is < 255
+pub(crate) fn sanity_checks(
+    index: &Index,
+    rtxn: &heed::RoTxn,
+    field_id: FieldId,
+    facet_type: FacetType,
+    group_size: usize,
+    _min_level_size: usize, // might add a check on level size later
+    max_group_size: usize,
+) -> Result<()> {
+    tracing::info!(%field_id, ?facet_type, "performing sanity checks");
+    let database = match facet_type {
+        FacetType::String => {
+            index.facet_id_string_docids.remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>()
+        }
+        FacetType::Number => {
+            index.facet_id_f64_docids.remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>()
+        }
+    };
+
+    let leaf_prefix: FacetGroupKey<&[u8]> = FacetGroupKey { field_id, level: 0, left_bound: &[] };
+
+    let leaf_it = database.prefix_iter(rtxn, &leaf_prefix)?;
+
+    let max_level = get_highest_level(rtxn, database, field_id)?;
+    if max_level == u8::MAX {
+        panic!("max_level == 255");
+    }
+
+    for leaf in leaf_it {
+        let (leaf_facet_value, leaf_docids) = leaf?;
+        let mut current_level = 0;
+
+        let mut current_parent_facet_value: Option<FacetGroupKey<&[u8]>> = None;
+        let mut current_parent_docids: Option<crate::heed_codec::facet::FacetGroupValue> = None;
+        loop {
+            current_level += 1;
+            if current_level >= max_level {
+                break;
+            }
+            let parent_key_right_bound = FacetGroupKey {
+                field_id,
+                level: current_level,
+                left_bound: leaf_facet_value.left_bound,
+            };
+            let (parent_facet_value, parent_docids) = database
+                .get_lower_than_or_equal_to(rtxn, &parent_key_right_bound)?
+                .expect("no parent found");
+            if parent_facet_value.level != current_level {
+                panic!(
+                    "wrong parent level, found_level={}, expected_level={}",
+                    parent_facet_value.level, current_level
+                );
+            }
+            if parent_facet_value.field_id != field_id {
+                panic!("wrong parent fid");
+            }
+            if parent_facet_value.left_bound > leaf_facet_value.left_bound {
+                panic!("wrong parent left bound");
+            }
+
+            if !leaf_docids.bitmap.is_subset(&parent_docids.bitmap) {
+                panic!(
+                    "missing docids from leaf in parent, current_level={}, parent={}, child={}, missing={missing:?}, child_len={}, child={:?}",
+                    current_level,
+                    facet_to_string(parent_facet_value.left_bound, facet_type),
+                    facet_to_string(leaf_facet_value.left_bound, facet_type),
+                    leaf_docids.bitmap.len(),
+                    leaf_docids.bitmap.clone(),
+                    missing=leaf_docids.bitmap - parent_docids.bitmap,
+                )
+            }
+
+            if let Some(current_parent_facet_value) = current_parent_facet_value {
+                if current_parent_facet_value.field_id != parent_facet_value.field_id {
+                    panic!("wrong parent parent fid");
+                }
+                if current_parent_facet_value.level + 1 != parent_facet_value.level {
+                    panic!("wrong parent parent level");
+                }
+                if current_parent_facet_value.left_bound < parent_facet_value.left_bound {
+                    panic!("wrong parent parent left bound");
+                }
+            }
+
+            if let Some(current_parent_docids) = current_parent_docids {
+                if !current_parent_docids.bitmap.is_subset(&parent_docids.bitmap) {
+                    panic!("missing docids from intermediate node in parent, parent_level={}, parent={}, intermediate={}, missing={missing:?}, intermediate={:?}",
+                    parent_facet_value.level,
+                    facet_to_string(parent_facet_value.left_bound, facet_type),
+                    facet_to_string(current_parent_facet_value.unwrap().left_bound, facet_type),
+                    current_parent_docids.bitmap.clone(),
+                    missing=current_parent_docids.bitmap - parent_docids.bitmap,
+                    );
+                }
+            }
+
+            current_parent_facet_value = Some(parent_facet_value);
+            current_parent_docids = Some(parent_docids);
+        }
+    }
+    tracing::info!(%field_id, ?facet_type, "checked all leaves");
+
+    let mut current_level = max_level;
+    let mut greater_than_group = 0usize;
+    let mut total = 0usize;
+    loop {
+        if current_level == 0 {
+            break;
+        }
+        let child_level = current_level - 1;
+        tracing::info!(%field_id, ?facet_type, %current_level, "checked groups for level");
+        let level_groups_prefix: FacetGroupKey<&[u8]> =
+            FacetGroupKey { field_id, level: current_level, left_bound: &[] };
+        let mut level_groups_it = database.prefix_iter(rtxn, &level_groups_prefix)?.peekable();
+
+        'group_it: loop {
+            let Some(group) = level_groups_it.next() else { break 'group_it };
+
+            let (group_facet_value, group_docids) = group?;
+            let child_left_bound = group_facet_value.left_bound.to_owned();
+            let mut expected_docids = RoaringBitmap::new();
+            let mut expected_size = 0usize;
+            let right_bound = level_groups_it
+                .peek()
+                .and_then(|res| res.as_ref().ok())
+                .map(|(key, _)| key.left_bound);
+            let child_left_bound = FacetGroupKey {
+                field_id,
+                level: child_level,
+                left_bound: child_left_bound.as_slice(),
+            };
+            let child_left_bound = Bound::Included(&child_left_bound);
+            let child_right_bound;
+            let child_right_bound = if let Some(right_bound) = right_bound {
+                child_right_bound =
+                    FacetGroupKey { field_id, level: child_level, left_bound: right_bound };
+                Bound::Excluded(&child_right_bound)
+            } else {
+                Bound::Unbounded
+            };
+            let children = database.range(rtxn, &(child_left_bound, child_right_bound))?;
+            for child in children {
+                let (child_facet_value, child_docids) = child?;
+                if child_facet_value.field_id != field_id {
+                    break;
+                }
+                if child_facet_value.level != child_level {
+                    break;
+                }
+                expected_size += 1;
+                expected_docids |= &child_docids.bitmap;
+            }
+            assert_eq!(expected_size, group_docids.size as usize);
+            assert!(expected_size <= max_group_size);
+            assert_eq!(expected_docids, group_docids.bitmap);
+            total += 1;
+            if expected_size > group_size {
+                greater_than_group += 1;
+            }
+        }
+
+        current_level -= 1;
+    }
+    if greater_than_group * 2 > total {
+        panic!("too many groups have a size > group_size");
+    }
+
+    tracing::info!("sanity checks OK");
+
+    Ok(())
+}
+
+fn facet_to_string(facet_value: &[u8], facet_type: FacetType) -> String {
+    match facet_type {
+        FacetType::String => bstr::BStr::new(facet_value).to_string(),
+        FacetType::Number => match OrderedF64Codec::bytes_decode(facet_value) {
+            Ok(value) => value.to_string(),
+            Err(e) => format!("error: {e} (bytes: {facet_value:?}"),
+        },
+    }
+}

crates/milli/src/update/facet/new_incremental.rs

@@ -0,0 +1,498 @@
+use std::ops::Bound;
+
+use heed::types::{Bytes, DecodeIgnore};
+use heed::{BytesDecode as _, Database, RwTxn};
+use roaring::RoaringBitmap;
+
+use crate::facet::FacetType;
+use crate::heed_codec::facet::{
+    FacetGroupKey, FacetGroupKeyCodec, FacetGroupValue, FacetGroupValueCodec,
+};
+use crate::heed_codec::BytesRefCodec;
+use crate::search::facet::get_highest_level;
+use crate::update::valid_facet_value;
+use crate::{FieldId, Index, Result};
+
+pub struct FacetsUpdateIncremental {
+    inner: FacetsUpdateIncrementalInner,
+    delta_data: Vec<FacetFieldIdChange>,
+}
+
+struct FacetsUpdateIncrementalInner {
+    db: Database<FacetGroupKeyCodec<BytesRefCodec>, FacetGroupValueCodec>,
+    field_id: FieldId,
+    group_size: u8,
+    min_level_size: u8,
+    max_group_size: u8,
+}
+
+impl FacetsUpdateIncremental {
+    pub fn new(
+        index: &Index,
+        facet_type: FacetType,
+        field_id: FieldId,
+        delta_data: Vec<FacetFieldIdChange>,
+        group_size: u8,
+        min_level_size: u8,
+        max_group_size: u8,
+    ) -> Self {
+        FacetsUpdateIncremental {
+            inner: FacetsUpdateIncrementalInner {
+                db: match facet_type {
+                    FacetType::String => index
+                        .facet_id_string_docids
+                        .remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>(),
+                    FacetType::Number => index
+                        .facet_id_f64_docids
+                        .remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>(),
+                },
+                field_id,
+                group_size,
+                min_level_size,
+                max_group_size,
+            },
+
+            delta_data,
+        }
+    }
+
+    #[tracing::instrument(level = "trace", skip_all, target = "indexing::facets::incremental")]
+    pub fn execute(mut self, wtxn: &mut RwTxn) -> Result<()> {
+        if self.delta_data.is_empty() {
+            return Ok(());
+        }
+        self.delta_data.sort_unstable_by(
+            |FacetFieldIdChange { facet_value: left, .. },
+             FacetFieldIdChange { facet_value: right, .. }| {
+                left.cmp(right)
+                    // sort in **reverse** lexicographic order
+                    .reverse()
+            },
+        );
+
+        self.inner.find_changed_parents(wtxn, self.delta_data)?;
+
+        self.inner.add_or_delete_level(wtxn)
+    }
+}
+
+impl FacetsUpdateIncrementalInner {
+    /// WARNING: `changed_children` must be sorted in **reverse** lexicographic order.
+    fn find_changed_parents(
+        &self,
+        wtxn: &mut RwTxn,
+        mut changed_children: Vec<FacetFieldIdChange>,
+    ) -> Result<()> {
+        let mut changed_parents = vec![];
+        for child_level in 0u8..u8::MAX {
+            // child_level < u8::MAX by construction
+            let parent_level = child_level + 1;
+            let parent_level_left_bound: FacetGroupKey<&[u8]> =
+                FacetGroupKey { field_id: self.field_id, level: parent_level, left_bound: &[] };
+
+            let mut last_parent: Option<Box<[u8]>> = None;
+            let mut child_it = changed_children
+                // drain all changed children
+                .drain(..)
+                // keep only children whose value is valid in the LMDB sense
+                .filter(|child| valid_facet_value(&child.facet_value));
+            // `while let` rather than `for` because we advance `child_it` inside of the loop
+            'current_level: while let Some(child) = child_it.next() {
+                if let Some(last_parent) = &last_parent {
+                    if &child.facet_value >= last_parent {
+                        self.compute_parent_group(wtxn, child_level, child.facet_value)?;
+                        continue 'current_level;
+                    }
+                }
+
+                // need to find a new parent
+                let parent_key_prefix = FacetGroupKey {
+                    field_id: self.field_id,
+                    level: parent_level,
+                    left_bound: &*child.facet_value,
+                };
+
+                let parent = self
+                    .db
+                    .remap_data_type::<DecodeIgnore>()
+                    .rev_range(
+                        wtxn,
+                        &(
+                            Bound::Excluded(&parent_level_left_bound),
+                            Bound::Included(&parent_key_prefix),
+                        ),
+                    )?
+                    .next();
+
+                match parent {
+                    Some(Ok((parent_key, _parent_value))) => {
+                        // found parent, cache it for next keys
+                        last_parent = Some(parent_key.left_bound.to_owned().into_boxed_slice());
+
+                        // add to modified list for parent level
+                        changed_parents.push(FacetFieldIdChange {
+                            facet_value: parent_key.left_bound.to_owned().into_boxed_slice(),
+                        });
+                        self.compute_parent_group(wtxn, child_level, child.facet_value)?;
+                    }
+                    Some(Err(err)) => return Err(err.into()),
+                    None => {
+                        // no parent for that key
+                        let mut parent_it = self
+                            .db
+                            .remap_data_type::<DecodeIgnore>()
+                            .prefix_iter_mut(wtxn, &parent_level_left_bound)?;
+                        match parent_it.next() {
+                            // 1. left of the current left bound, or
+                            Some(Ok((first_key, _first_value))) => {
+                                // make sure we don't spill on the neighboring fid (level also included defensively)
+                                if first_key.field_id != self.field_id
+                                    || first_key.level != parent_level
+                                {
+                                    // max level reached, exit
+                                    drop(parent_it);
+                                    self.compute_parent_group(
+                                        wtxn,
+                                        child_level,
+                                        child.facet_value,
+                                    )?;
+                                    for child in child_it.by_ref() {
+                                        self.compute_parent_group(
+                                            wtxn,
+                                            child_level,
+                                            child.facet_value,
+                                        )?;
+                                    }
+                                    return Ok(());
+                                }
+                                // remove old left bound
+                                unsafe { parent_it.del_current()? };
+                                drop(parent_it);
+                                changed_parents.push(FacetFieldIdChange {
+                                    facet_value: child.facet_value.clone(),
+                                });
+                                self.compute_parent_group(wtxn, child_level, child.facet_value)?;
+                                // pop all elements in order to visit the new left bound
+                                let new_left_bound =
+                                    &mut changed_parents.last_mut().unwrap().facet_value;
+                                for child in child_it.by_ref() {
+                                    new_left_bound.clone_from(&child.facet_value);
+
+                                    self.compute_parent_group(
+                                        wtxn,
+                                        child_level,
+                                        child.facet_value,
+                                    )?;
+                                }
+                            }
+                            Some(Err(err)) => return Err(err.into()),
+                            // 2. max level reached, exit
+                            None => {
+                                drop(parent_it);
+                                self.compute_parent_group(wtxn, child_level, child.facet_value)?;
+                                for child in child_it.by_ref() {
+                                    self.compute_parent_group(
+                                        wtxn,
+                                        child_level,
+                                        child.facet_value,
+                                    )?;
+                                }
+                                return Ok(());
+                            }
+                        }
+                    }
+                }
+            }
+            if changed_parents.is_empty() {
+                return Ok(());
+            }
+            drop(child_it);
+            std::mem::swap(&mut changed_children, &mut changed_parents);
+            // changed_parents is now empty because changed_children was emptied by the drain
+        }
+        Ok(())
+    }
+
+    fn compute_parent_group(
+        &self,
+        wtxn: &mut RwTxn<'_>,
+        parent_level: u8,
+        parent_left_bound: Box<[u8]>,
+    ) -> Result<()> {
+        let mut range_left_bound: Vec<u8> = parent_left_bound.into();
+        if parent_level == 0 {
+            return Ok(());
+        }
+        let child_level = parent_level - 1;
+
+        let parent_key = FacetGroupKey {
+            field_id: self.field_id,
+            level: parent_level,
+            left_bound: &*range_left_bound,
+        };
+        let child_right_bound = self
+            .db
+            .remap_data_type::<DecodeIgnore>()
+            .get_greater_than(wtxn, &parent_key)?
+            .and_then(
+                |(
+                    FacetGroupKey {
+                        level: right_level,
+                        field_id: right_fid,
+                        left_bound: right_bound,
+                    },
+                    _,
+                )| {
+                    if parent_level != right_level || self.field_id != right_fid {
+                        // there was a greater key, but with a greater level or fid, so not a sibling to the parent: ignore
+                        return None;
+                    }
+                    Some(right_bound.to_owned())
+                },
+            );
+        let child_right_bound = match &child_right_bound {
+            Some(right_bound) => Bound::Excluded(FacetGroupKey {
+                left_bound: right_bound.as_slice(),
+                field_id: self.field_id,
+                level: child_level,
+            }),
+            None => Bound::Unbounded,
+        };
+
+        let child_left_key = FacetGroupKey {
+            field_id: self.field_id,
+            level: child_level,
+            left_bound: &*range_left_bound,
+        };
+        let mut child_left_bound = Bound::Included(child_left_key);
+
+        loop {
+            // do a first pass on the range to find the number of children
+            let child_count = self
+                .db
+                .remap_data_type::<DecodeIgnore>()
+                .range(wtxn, &(child_left_bound, child_right_bound))?
+                .take(self.max_group_size as usize * 2)
+                .count();
+            let mut child_it = self.db.range(wtxn, &(child_left_bound, child_right_bound))?;
+
+            // pick the right group_size depending on the number of children
+            let group_size = if child_count >= self.max_group_size as usize * 2 {
+                // more than twice the max_group_size => there will be space for at least 2 groups of max_group_size
+                self.max_group_size as usize
+            } else if child_count >= self.group_size as usize {
+                // size in [group_size, max_group_size * 2[
+                // divided by 2 it is between [group_size / 2, max_group_size[
+                // this ensures that the tree is balanced
+                child_count / 2
+            } else {
+                // take everything
+                child_count
+            };
+
+            let res: Result<_> = child_it
+                .by_ref()
+                .take(group_size)
+                // stop if we go to the next level or field id
+                .take_while(|res| match res {
+                    Ok((child_key, _)) => {
+                        child_key.field_id == self.field_id && child_key.level == child_level
+                    }
+                    Err(_) => true,
+                })
+                .try_fold(
+                    (None, FacetGroupValue { size: 0, bitmap: Default::default() }),
+                    |(bounds, mut group_value), child_res| {
+                        let (child_key, child_value) = child_res?;
+                        let bounds = match bounds {
+                            Some((left_bound, _)) => Some((left_bound, child_key.left_bound)),
+                            None => Some((child_key.left_bound, child_key.left_bound)),
+                        };
+                        // max_group_size <= u8::MAX
+                        group_value.size += 1;
+                        group_value.bitmap |= &child_value.bitmap;
+                        Ok((bounds, group_value))
+                    },
+                );
+
+            let (bounds, group_value) = res?;
+
+            let Some((group_left_bound, right_bound)) = bounds else {
+                let update_key = FacetGroupKey {
+                    field_id: self.field_id,
+                    level: parent_level,
+                    left_bound: &*range_left_bound,
+                };
+                drop(child_it);
+                if let Bound::Included(_) = child_left_bound {
+                    self.db.delete(wtxn, &update_key)?;
+                }
+
+                break;
+            };
+
+            drop(child_it);
+            let current_left_bound = group_left_bound.to_owned();
+
+            let delete_old_bound = match child_left_bound {
+                Bound::Included(bound) => {
+                    if bound.left_bound != current_left_bound {
+                        Some(range_left_bound.clone())
+                    } else {
+                        None
+                    }
+                }
+                _ => None,
+            };
+
+            range_left_bound.clear();
+            range_left_bound.extend_from_slice(right_bound);
+            let child_left_key = FacetGroupKey {
+                field_id: self.field_id,
+                level: child_level,
+                left_bound: range_left_bound.as_slice(),
+            };
+            child_left_bound = Bound::Excluded(child_left_key);
+
+            if let Some(old_bound) = delete_old_bound {
+                let update_key = FacetGroupKey {
+                    field_id: self.field_id,
+                    level: parent_level,
+                    left_bound: old_bound.as_slice(),
+                };
+                self.db.delete(wtxn, &update_key)?;
+            }
+
+            let update_key = FacetGroupKey {
+                field_id: self.field_id,
+                level: parent_level,
+                left_bound: current_left_bound.as_slice(),
+            };
+            if group_value.bitmap.is_empty() {
+                self.db.delete(wtxn, &update_key)?;
+            } else {
+                self.db.put(wtxn, &update_key, &group_value)?;
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Check whether the highest level has exceeded `min_level_size` * `self.group_size`.
+    /// If it has, we must build an addition level above it.
+    /// Then check whether the highest level is under `min_level_size`.
+    /// If it has, we must remove the complete level.
+    pub(crate) fn add_or_delete_level(&self, txn: &mut RwTxn<'_>) -> Result<()> {
+        let highest_level = get_highest_level(txn, self.db, self.field_id)?;
+        let mut highest_level_prefix = vec![];
+        highest_level_prefix.extend_from_slice(&self.field_id.to_be_bytes());
+        highest_level_prefix.push(highest_level);
+
+        let size_highest_level =
+            self.db.remap_types::<Bytes, Bytes>().prefix_iter(txn, &highest_level_prefix)?.count();
+
+        if size_highest_level >= self.group_size as usize * self.min_level_size as usize {
+            self.add_level(txn, highest_level, &highest_level_prefix, size_highest_level)
+        } else if size_highest_level < self.min_level_size as usize && highest_level != 0 {
+            self.delete_level(txn, &highest_level_prefix)
+        } else {
+            Ok(())
+        }
+    }
+
+    /// Delete a level.
+    fn delete_level(&self, txn: &mut RwTxn<'_>, highest_level_prefix: &[u8]) -> Result<()> {
+        let mut to_delete = vec![];
+        let mut iter =
+            self.db.remap_types::<Bytes, Bytes>().prefix_iter(txn, highest_level_prefix)?;
+        for el in iter.by_ref() {
+            let (k, _) = el?;
+            to_delete.push(
+                FacetGroupKeyCodec::<BytesRefCodec>::bytes_decode(k)
+                    .map_err(heed::Error::Encoding)?
+                    .into_owned(),
+            );
+        }
+        drop(iter);
+        for k in to_delete {
+            self.db.delete(txn, &k.as_ref())?;
+        }
+        Ok(())
+    }
+
+    /// Build an additional level for the field id.
+    fn add_level(
+        &self,
+        txn: &mut RwTxn<'_>,
+        highest_level: u8,
+        highest_level_prefix: &[u8],
+        size_highest_level: usize,
+    ) -> Result<()> {
+        let mut groups_iter = self
+            .db
+            .remap_types::<Bytes, FacetGroupValueCodec>()
+            .prefix_iter(txn, highest_level_prefix)?;
+
+        let nbr_new_groups = size_highest_level / self.group_size as usize;
+        let nbr_leftover_elements = size_highest_level % self.group_size as usize;
+
+        let mut to_add = vec![];
+        for _ in 0..nbr_new_groups {
+            let mut first_key = None;
+            let mut values = RoaringBitmap::new();
+            for _ in 0..self.group_size {
+                let (key_bytes, value_i) = groups_iter.next().unwrap()?;
+                let key_i = FacetGroupKeyCodec::<BytesRefCodec>::bytes_decode(key_bytes)
+                    .map_err(heed::Error::Encoding)?;
+
+                if first_key.is_none() {
+                    first_key = Some(key_i);
+                }
+                values |= value_i.bitmap;
+            }
+            let key = FacetGroupKey {
+                field_id: self.field_id,
+                level: highest_level + 1,
+                left_bound: first_key.unwrap().left_bound,
+            };
+            let value = FacetGroupValue { size: self.group_size, bitmap: values };
+            to_add.push((key.into_owned(), value));
+        }
+        // now we add the rest of the level, in case its size is > group_size * min_level_size
+        // this can indeed happen if the min_level_size parameter changes between two calls to `insert`
+        if nbr_leftover_elements > 0 {
+            let mut first_key = None;
+            let mut values = RoaringBitmap::new();
+            for _ in 0..nbr_leftover_elements {
+                let (key_bytes, value_i) = groups_iter.next().unwrap()?;
+                let key_i = FacetGroupKeyCodec::<BytesRefCodec>::bytes_decode(key_bytes)
+                    .map_err(heed::Error::Encoding)?;
+
+                if first_key.is_none() {
+                    first_key = Some(key_i);
+                }
+                values |= value_i.bitmap;
+            }
+            let key = FacetGroupKey {
+                field_id: self.field_id,
+                level: highest_level + 1,
+                left_bound: first_key.unwrap().left_bound,
+            };
+            // Note: nbr_leftover_elements can be casted to a u8 since it is bounded by `max_group_size`
+            // when it is created above.
+            let value = FacetGroupValue { size: nbr_leftover_elements as u8, bitmap: values };
+            to_add.push((key.into_owned(), value));
+        }
+
+        drop(groups_iter);
+        for (key, value) in to_add {
+            self.db.put(txn, &key.as_ref(), &value)?;
+        }
+        Ok(())
+    }
+}
+
+#[derive(Debug)]
+pub struct FacetFieldIdChange {
+    pub facet_value: Box<[u8]>,
+}

crates/milli/src/update/index_documents/helpers/mod.rs

@@ -10,10 +10,14 @@ use fst::{IntoStreamer, Streamer};
 pub use grenad_helpers::*;
 pub use merge_functions::*;
 
-use crate::MAX_WORD_LENGTH;
+use crate::MAX_LMDB_KEY_LENGTH;
 
 pub fn valid_lmdb_key(key: impl AsRef<[u8]>) -> bool {
-    key.as_ref().len() <= MAX_WORD_LENGTH * 2 && !key.as_ref().is_empty()
+    key.as_ref().len() <= MAX_LMDB_KEY_LENGTH - 3 && !key.as_ref().is_empty()
+}
+
+pub fn valid_facet_value(facet_value: impl AsRef<[u8]>) -> bool {
+    facet_value.as_ref().len() <= MAX_LMDB_KEY_LENGTH - 3 && !facet_value.as_ref().is_empty()
 }
 
 /// Divides one slice into two at an index, returns `None` if mid is out of bounds.

crates/milli/src/update/index_documents/mod.rs

@@ -3334,6 +3334,44 @@ mod tests {
         rtxn.commit().unwrap();
     }
 
+    #[test]
+    fn incremental_update_without_changing_facet_distribution() {
+        let index = TempIndex::new();
+        index
+            .add_documents(documents!([
+                {"id": 0, "some_field": "aaa", "other_field": "aaa" },
+                {"id": 1, "some_field": "bbb", "other_field": "bbb" },
+            ]))
+            .unwrap();
+        {
+            let rtxn = index.read_txn().unwrap();
+            // count field distribution
+            let results = index.field_distribution(&rtxn).unwrap();
+            assert_eq!(Some(&2), results.get("id"));
+            assert_eq!(Some(&2), results.get("some_field"));
+            assert_eq!(Some(&2), results.get("other_field"));
+        }
+
+        let mut index = index;
+        index.index_documents_config.update_method = IndexDocumentsMethod::UpdateDocuments;
+
+        index
+            .add_documents(documents!([
+                {"id": 0, "other_field": "bbb" },
+                {"id": 1, "some_field": "ccc" },
+            ]))
+            .unwrap();
+
+        {
+            let rtxn = index.read_txn().unwrap();
+            // count field distribution
+            let results = index.field_distribution(&rtxn).unwrap();
+            assert_eq!(Some(&2), results.get("id"));
+            assert_eq!(Some(&2), results.get("some_field"));
+            assert_eq!(Some(&2), results.get("other_field"));
+        }
+    }
+
     #[test]
     fn delete_words_exact_attributes() {
         let index = TempIndex::new();

crates/milli/src/update/index_documents/snapshots/mod.rs/geo_filtered_placeholder_search_should_not_return_deleted_documents/facet_id_f64_docids.snap

@@ -1,5 +1,5 @@
 ---
-source: milli/src/update/index_documents/mod.rs
+source: crates/milli/src/update/index_documents/mod.rs
 ---
 3   0  48.9021 1  [19, ]
 3   0  49.9314 1  [17, ]
@@ -15,6 +15,11 @@ source: milli/src/update/index_documents/mod.rs
 3   0  50.7453 1  [7, ]
 3   0  50.8466 1  [10, ]
 3   0  51.0537 1  [9, ]
+3   1  48.9021 2  [17, 19, ]
+3   1  50.1793 3  [13, 14, 15, ]
+3   1  50.4502 4  [0, 3, 8, 12, ]
+3   1  50.6312 2  [1, 2, ]
+3   1  50.7453 3  [7, 9, 10, ]
 4   0  2.271  1  [17, ]
 4   0  2.3708 1  [19, ]
 4   0  2.7637 1  [14, ]
@@ -28,4 +33,3 @@ source: milli/src/update/index_documents/mod.rs
 4   0  3.6957 1  [9, ]
 4   0  3.9623 1  [12, ]
 4   0  4.337  1  [10, ]
-

crates/milli/src/update/new/extract/documents.rs

@@ -89,7 +89,8 @@ impl<'a, 'b, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a, 'b> {
                             .or_default();
                         *entry -= 1;
                     }
-                    let content = update.updated();
+                    let content =
+                        update.merged(&context.rtxn, context.index, &context.db_fields_ids_map)?;
                     let geo_iter =
                         content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
                     for res in content.iter_top_level_fields().chain(geo_iter) {

crates/milli/src/update/new/indexer/document_operation.rs

@@ -252,6 +252,24 @@ fn extract_addition_payload_changes<'r, 'pl: 'r>(
         previous_offset = iter.byte_offset();
     }
 
+    if payload.is_empty() {
+        let result = retrieve_or_guess_primary_key(
+            rtxn,
+            index,
+            new_fields_ids_map,
+            primary_key_from_op,
+            None,
+        );
+        match result {
+            Ok(Ok((pk, _))) => {
+                primary_key.get_or_insert(pk);
+            }
+            Ok(Err(UserError::NoPrimaryKeyCandidateFound)) => (),
+            Ok(Err(user_error)) => return Err(Error::UserError(user_error)),
+            Err(error) => return Err(error),
+        };
+    }
+
     Ok(new_docids_version_offsets)
 }
 

crates/milli/src/update/new/indexer/mod.rs

@@ -4,6 +4,7 @@ use std::sync::{OnceLock, RwLock};
 use std::thread::{self, Builder};
 
 use big_s::S;
+use bstr::ByteSlice as _;
 use bumparaw_collections::RawMap;
 use document_changes::{extract, DocumentChanges, IndexingContext};
 pub use document_deletion::DocumentDeletion;
@@ -36,6 +37,8 @@ use crate::index::main_key::{WORDS_FST_KEY, WORDS_PREFIXES_FST_KEY};
 use crate::progress::Progress;
 use crate::proximity::ProximityPrecision;
 use crate::update::del_add::DelAdd;
+use crate::update::facet::new_incremental::FacetsUpdateIncremental;
+use crate::update::facet::{FACET_GROUP_SIZE, FACET_MAX_GROUP_SIZE, FACET_MIN_LEVEL_SIZE};
 use crate::update::new::extract::EmbeddingExtractor;
 use crate::update::new::merger::merge_and_send_rtree;
 use crate::update::new::words_prefix_docids::compute_exact_word_prefix_docids;
@@ -107,7 +110,7 @@ where
         },
     );
 
-    let (extractor_sender, mut writer_receiver) = pool
+    let (extractor_sender, writer_receiver) = pool
         .install(|| extractor_writer_bbqueue(&mut bbbuffers, total_bbbuffer_capacity, 1000))
         .unwrap();
 
@@ -203,6 +206,7 @@ where
                             caches,
                             FacetDatabases::new(index),
                             index,
+                            &rtxn,
                             extractor_sender.facet_docids(),
                         )?;
                     }
@@ -421,6 +425,7 @@ where
         let mut arroy_writers = arroy_writers?;
 
         {
+            let mut writer_receiver = writer_receiver;
             let span = tracing::trace_span!(target: "indexing::write_db", "all");
             let _entered = span.enter();
 
@@ -580,7 +585,12 @@ fn write_from_bbqueue(
                     }
                     (key, None) => match database.delete(wtxn, key) {
                         Ok(false) => {
-                            unreachable!("We tried to delete an unknown key: {key:?}")
+                            tracing::error!(
+                                database_name,
+                                key_bytes = ?key,
+                                formatted_key = ?key.as_bstr(),
+                                "Attempt to delete an unknown key"
+                            );
                         }
                         Ok(_) => (),
                         Err(error) => {
@@ -735,27 +745,66 @@ fn compute_facet_search_database(
 fn compute_facet_level_database(
     index: &Index,
     wtxn: &mut RwTxn,
-    facet_field_ids_delta: FacetFieldIdsDelta,
+    mut facet_field_ids_delta: FacetFieldIdsDelta,
 ) -> Result<()> {
-    if let Some(modified_facet_string_ids) = facet_field_ids_delta.modified_facet_string_ids() {
+    for (fid, delta) in facet_field_ids_delta.consume_facet_string_delta() {
         let span = tracing::trace_span!(target: "indexing::facet_field_ids", "string");
         let _entered = span.enter();
-        FacetsUpdateBulk::new_not_updating_level_0(
+        match delta {
-            index,
+            super::merger::FacetFieldIdDelta::Bulk => {
-            modified_facet_string_ids,
+                tracing::debug!(%fid, "bulk string facet processing");
-            FacetType::String,
+                FacetsUpdateBulk::new_not_updating_level_0(index, vec![fid], FacetType::String)
-        )
+                    .execute(wtxn)?
-        .execute(wtxn)?;
+            }
+            super::merger::FacetFieldIdDelta::Incremental(delta_data) => {
+                tracing::debug!(%fid, len=%delta_data.len(), "incremental string facet processing");
+                FacetsUpdateIncremental::new(
+                    index,
+                    FacetType::String,
+                    fid,
+                    delta_data,
+                    FACET_GROUP_SIZE,
+                    FACET_MIN_LEVEL_SIZE,
+                    FACET_MAX_GROUP_SIZE,
+                )
+                .execute(wtxn)?
+            }
+        }
     }
-    if let Some(modified_facet_number_ids) = facet_field_ids_delta.modified_facet_number_ids() {
+
+    for (fid, delta) in facet_field_ids_delta.consume_facet_number_delta() {
         let span = tracing::trace_span!(target: "indexing::facet_field_ids", "number");
         let _entered = span.enter();
-        FacetsUpdateBulk::new_not_updating_level_0(
+        match delta {
+            super::merger::FacetFieldIdDelta::Bulk => {
+                tracing::debug!(%fid, "bulk number facet processing");
+                FacetsUpdateBulk::new_not_updating_level_0(index, vec![fid], FacetType::Number)
+                    .execute(wtxn)?
+            }
+            super::merger::FacetFieldIdDelta::Incremental(delta_data) => {
+                tracing::debug!(%fid, len=%delta_data.len(), "incremental number facet processing");
+                FacetsUpdateIncremental::new(
+                    index,
+                    FacetType::Number,
+                    fid,
+                    delta_data,
+                    FACET_GROUP_SIZE,
+                    FACET_MIN_LEVEL_SIZE,
+                    FACET_MAX_GROUP_SIZE,
+                )
+                .execute(wtxn)?
+            }
+        }
+        debug_assert!(crate::update::facet::sanity_checks(
             index,
-            modified_facet_number_ids,
+            wtxn,
+            fid,
             FacetType::Number,
+            FACET_GROUP_SIZE as usize,
+            FACET_MIN_LEVEL_SIZE as usize,
+            FACET_MAX_GROUP_SIZE as usize,
         )
-        .execute(wtxn)?;
+        .is_ok());
     }
 
     Ok(())

crates/milli/src/update/new/merger.rs

@@ -1,6 +1,6 @@
 use std::cell::RefCell;
 
-use hashbrown::HashSet;
+use hashbrown::HashMap;
 use heed::types::Bytes;
 use heed::{Database, RoTxn};
 use memmap2::Mmap;
@@ -12,6 +12,7 @@ use super::extract::{
     merge_caches_sorted, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap,
     FacetKind, GeoExtractorData,
 };
+use crate::update::facet::new_incremental::FacetFieldIdChange;
 use crate::{CboRoaringBitmapCodec, FieldId, GeoPoint, Index, InternalError, Result};
 
 #[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
@@ -100,12 +101,18 @@ pub fn merge_and_send_facet_docids<'extractor>(
     mut caches: Vec<BalancedCaches<'extractor>>,
     database: FacetDatabases,
     index: &Index,
+    rtxn: &RoTxn,
     docids_sender: FacetDocidsSender,
 ) -> Result<FacetFieldIdsDelta> {
+    let max_string_count = (index.facet_id_string_docids.len(rtxn)? / 500) as usize;
+    let max_number_count = (index.facet_id_f64_docids.len(rtxn)? / 500) as usize;
+    let max_string_count = max_string_count.clamp(1000, 100_000);
+    let max_number_count = max_number_count.clamp(1000, 100_000);
     transpose_and_freeze_caches(&mut caches)?
         .into_par_iter()
         .map(|frozen| {
-            let mut facet_field_ids_delta = FacetFieldIdsDelta::default();
+            let mut facet_field_ids_delta =
+                FacetFieldIdsDelta::new(max_string_count, max_number_count);
             let rtxn = index.read_txn()?;
             merge_caches_sorted(frozen, |key, DelAddRoaringBitmap { del, add }| {
                 let current = database.get_cbo_roaring_bytes_value(&rtxn, key)?;
@@ -126,7 +133,10 @@ pub fn merge_and_send_facet_docids<'extractor>(
 
             Ok(facet_field_ids_delta)
         })
-        .reduce(|| Ok(FacetFieldIdsDelta::default()), |lhs, rhs| Ok(lhs?.merge(rhs?)))
+        .reduce(
+            || Ok(FacetFieldIdsDelta::new(max_string_count, max_number_count)),
+            |lhs, rhs| Ok(lhs?.merge(rhs?)),
+        )
 }
 
 pub struct FacetDatabases<'a> {
@@ -155,60 +165,131 @@ impl<'a> FacetDatabases<'a> {
     }
 }
 
-#[derive(Debug, Default)]
+#[derive(Debug)]
+pub enum FacetFieldIdDelta {
+    Bulk,
+    Incremental(Vec<FacetFieldIdChange>),
+}
+
+impl FacetFieldIdDelta {
+    fn push(&mut self, facet_value: &[u8], max_count: usize) {
+        *self = match std::mem::replace(self, FacetFieldIdDelta::Bulk) {
+            FacetFieldIdDelta::Bulk => FacetFieldIdDelta::Bulk,
+            FacetFieldIdDelta::Incremental(mut v) => {
+                if v.len() >= max_count {
+                    FacetFieldIdDelta::Bulk
+                } else {
+                    v.push(FacetFieldIdChange { facet_value: facet_value.into() });
+                    FacetFieldIdDelta::Incremental(v)
+                }
+            }
+        }
+    }
+
+    fn merge(&mut self, rhs: Option<Self>, max_count: usize) {
+        let Some(rhs) = rhs else {
+            return;
+        };
+        *self = match (std::mem::replace(self, FacetFieldIdDelta::Bulk), rhs) {
+            (FacetFieldIdDelta::Bulk, _) | (_, FacetFieldIdDelta::Bulk) => FacetFieldIdDelta::Bulk,
+            (
+                FacetFieldIdDelta::Incremental(mut left),
+                FacetFieldIdDelta::Incremental(mut right),
+            ) => {
+                if left.len() + right.len() >= max_count {
+                    FacetFieldIdDelta::Bulk
+                } else {
+                    left.append(&mut right);
+                    FacetFieldIdDelta::Incremental(left)
+                }
+            }
+        };
+    }
+}
+
+#[derive(Debug)]
 pub struct FacetFieldIdsDelta {
     /// The field ids that have been modified
-    modified_facet_string_ids: HashSet<FieldId>,
+    modified_facet_string_ids: HashMap<FieldId, FacetFieldIdDelta, rustc_hash::FxBuildHasher>,
-    modified_facet_number_ids: HashSet<FieldId>,
+    modified_facet_number_ids: HashMap<FieldId, FacetFieldIdDelta, rustc_hash::FxBuildHasher>,
+    max_string_count: usize,
+    max_number_count: usize,
 }
 
 impl FacetFieldIdsDelta {
-    fn register_facet_string_id(&mut self, field_id: FieldId) {
+    pub fn new(max_string_count: usize, max_number_count: usize) -> Self {
-        self.modified_facet_string_ids.insert(field_id);
+        Self {
+            max_string_count,
+            max_number_count,
+            modified_facet_string_ids: Default::default(),
+            modified_facet_number_ids: Default::default(),
+        }
     }
 
-    fn register_facet_number_id(&mut self, field_id: FieldId) {
+    fn register_facet_string_id(&mut self, field_id: FieldId, facet_value: &[u8]) {
-        self.modified_facet_number_ids.insert(field_id);
+        self.modified_facet_string_ids
+            .entry(field_id)
+            .or_insert(FacetFieldIdDelta::Incremental(Default::default()))
+            .push(facet_value, self.max_string_count);
+    }
+
+    fn register_facet_number_id(&mut self, field_id: FieldId, facet_value: &[u8]) {
+        self.modified_facet_number_ids
+            .entry(field_id)
+            .or_insert(FacetFieldIdDelta::Incremental(Default::default()))
+            .push(facet_value, self.max_number_count);
     }
 
     fn register_from_key(&mut self, key: &[u8]) {
-        let (facet_kind, field_id) = self.extract_key_data(key);
+        let (facet_kind, field_id, facet_value) = self.extract_key_data(key);
-        match facet_kind {
+        match (facet_kind, facet_value) {
-            FacetKind::Number => self.register_facet_number_id(field_id),
+            (FacetKind::Number, Some(facet_value)) => {
-            FacetKind::String => self.register_facet_string_id(field_id),
+                self.register_facet_number_id(field_id, facet_value)
+            }
+            (FacetKind::String, Some(facet_value)) => {
+                self.register_facet_string_id(field_id, facet_value)
+            }
             _ => (),
         }
     }
 
-    fn extract_key_data(&self, key: &[u8]) -> (FacetKind, FieldId) {
+    fn extract_key_data<'key>(&self, key: &'key [u8]) -> (FacetKind, FieldId, Option<&'key [u8]>) {
         let facet_kind = FacetKind::from(key[0]);
         let field_id = FieldId::from_be_bytes([key[1], key[2]]);
-        (facet_kind, field_id)
+        let facet_value = if key.len() >= 4 {
+            // level is also stored in the key at [3] (always 0)
+            Some(&key[4..])
+        } else {
+            None
+        };
+
+        (facet_kind, field_id, facet_value)
     }
 
-    pub fn modified_facet_string_ids(&self) -> Option<Vec<FieldId>> {
+    pub fn consume_facet_string_delta(
-        if self.modified_facet_string_ids.is_empty() {
+        &mut self,
-            None
+    ) -> impl Iterator<Item = (FieldId, FacetFieldIdDelta)> + '_ {
-        } else {
+        self.modified_facet_string_ids.drain()
-            Some(self.modified_facet_string_ids.iter().copied().collect())
-        }
     }
 
-    pub fn modified_facet_number_ids(&self) -> Option<Vec<FieldId>> {
+    pub fn consume_facet_number_delta(
-        if self.modified_facet_number_ids.is_empty() {
+        &mut self,
-            None
+    ) -> impl Iterator<Item = (FieldId, FacetFieldIdDelta)> + '_ {
-        } else {
+        self.modified_facet_number_ids.drain()
-            Some(self.modified_facet_number_ids.iter().copied().collect())
-        }
     }
 
     pub fn merge(mut self, rhs: Self) -> Self {
-        let Self { modified_facet_number_ids, modified_facet_string_ids } = rhs;
+        // rhs.max_xx_count is assumed to be equal to self.max_xx_count, and so gets unused
-        modified_facet_number_ids.into_iter().for_each(|fid| {
+        let Self { modified_facet_number_ids, modified_facet_string_ids, .. } = rhs;
-            self.modified_facet_number_ids.insert(fid);
+        modified_facet_number_ids.into_iter().for_each(|(fid, mut delta)| {
+            let old_delta = self.modified_facet_number_ids.remove(&fid);
+            delta.merge(old_delta, self.max_number_count);
+            self.modified_facet_number_ids.insert(fid, delta);
         });
-        modified_facet_string_ids.into_iter().for_each(|fid| {
+        modified_facet_string_ids.into_iter().for_each(|(fid, mut delta)| {
-            self.modified_facet_string_ids.insert(fid);
+            let old_delta = self.modified_facet_string_ids.remove(&fid);
+            delta.merge(old_delta, self.max_string_count);
+            self.modified_facet_string_ids.insert(fid, delta);
         });
         self
     }

crates/milli/src/update/new/mod.rs

@@ -16,6 +16,7 @@ pub mod indexer;
 mod merger;
 mod parallel_iterator_ext;
 mod ref_cell_ext;
+pub mod reindex;
 pub(crate) mod steps;
 pub(crate) mod thread_local;
 pub mod vector_document;

crates/milli/src/update/new/reindex.rs

@@ -0,0 +1,38 @@
+use heed::RwTxn;
+
+use super::document::{Document, DocumentFromDb};
+use crate::progress::{self, AtomicSubStep, Progress};
+use crate::{FieldDistribution, Index, Result};
+
+pub fn field_distribution(index: &Index, wtxn: &mut RwTxn<'_>, progress: &Progress) -> Result<()> {
+    let mut distribution = FieldDistribution::new();
+
+    let document_count = index.number_of_documents(wtxn)?;
+    let field_id_map = index.fields_ids_map(wtxn)?;
+
+    let (update_document_count, sub_step) =
+        AtomicSubStep::<progress::Document>::new(document_count as u32);
+    progress.update_progress(sub_step);
+
+    let docids = index.documents_ids(wtxn)?;
+
+    for docid in docids {
+        update_document_count.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
+
+        let Some(document) = DocumentFromDb::new(docid, wtxn, index, &field_id_map)? else {
+            continue;
+        };
+        let geo_iter = document.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
+        for res in document.iter_top_level_fields().chain(geo_iter) {
+            let (field_name, _) = res?;
+            if let Some(count) = distribution.get_mut(field_name) {
+                *count += 1;
+            } else {
+                distribution.insert(field_name.to_owned(), 1);
+            }
+        }
+    }
+
+    index.put_field_distribution(wtxn, &distribution)?;
+    Ok(())
+}